▪ RESEARCH · AI AGENTS · 2026 EDITION
Agents get wallets.
The interface layer of crypto is being replaced. Autonomous agents now hold wallets, pay each other per-request, trade NFTs, and read markets through APIs instead of screens. Web2 agents relied on clickstreams; Web3 agents need behavioral intelligence.
NEW USER CLASS
AUTONOMOUS AGENTS
PAYMENTS
MACHINE-NATIVE
MOAT
MEASURED BEHAVIOR
EXECUTIVE SUMMARY
Crypto’s original user was a human with a wallet. Its fastest-growing user class is software with a wallet. Agents hold keys, negotiate machine payments, mint and trade assets, and increasingly set the marginal price in thin markets. The bottleneck is no longer capability — it is trust and legibility: an agent transacting with an unknown wallet needs the same question answered that retail needed in 2017. Is this counterparty smart money or slop? That question is exactly what WalletDNA answers in machine-readable form.
SECTION 01
The Machine Payment Stack
FIG. 01AGENT MESH · WALLETS, PAYMENTS, MEASUREMENT
| LAYER | WHAT IT DOES | STATE IN 2026 |
|---|---|---|
| Agent wallets (ERC-4337 / session keys) | Provisioned, gasless, scoped-permission accounts | Production — the default agent chassis |
| Per-request payment rails (x402-style) | HTTP-native micropayments for API/content access | Adopting — replaces subscriptions for machine consumption |
| Agentic market access | DEX aggregators + RFQ + event contracts via API | Mainstream for market-making agents |
| Provenance registries | Signed provenance for generated media + datasets | Enterprise traction, consumer invisible |
| Agent-to-agent escrow | On-chain settlement of multi-step autonomous work | Early — the emerging primitive |
SECTION 02
Agentic NFTs & Ownership of Output
The NFT re-emerges here as output rights infrastructure: agents mint dynamic NFTs representing proprietary datasets, generated media and service agreements; ERC-6551 accounts bundle an agent’s whole inventory into one transferable object. Combined with our NFT thesis, the pattern is clear — the token isn’t the collectible, it’s the contract of ownership over machine work. Provenance matters more, not less, in a synthetic-media economy.
SECTION 03
Intelligence as the Moat
Every agent reads the same models. Differentiation comes from what they can measure: raw price feeds are commodities, but behavioral state — which wallets accumulate, which churn, which enter early — is proprietary when you compute it fresh at every block. This is the thesis behind WalletDNA as an agent-facing API: skill filters (isSmartMoney, isLoyalHolder, isEarlyEntrant, isWhale, isChurner, isExiting) that let any agent reason about counterparties in one GET. If Web2 agents ran on clickstreams, Web3 agents run on measured on-chain behavior — and the platforms that expose it own the intelligence layer of agentic commerce.
Models are equal. Measurements are not. The agent economy will price access to state, not access to intelligence.
PNTHR RESEARCH DESK · AI THESIS
THESIS IN ONE LINE
Agents don’t need dashboards — they need boolean truth about counterparties. Expose it as an endpoint and you become part of their stack. GET /api/wallet/dna
PNTHR AI RESEARCH DESK · AI THESIS, 2026 EDITION · OPINION WITH RECEIPTS · NOT FINANCIAL ADVICE · ALL THESES →